Efficient Approaches for Identifying Device with Short Retention Time in 6F2 DRAM Cells with Geometric Fluctuation

Abstract

Over the past 30 years, the size of Dynamic Random Access Memory (DRAM) has continuously decreased. Controlling data retention time is crucial, particularly with the ongoing reduction in device size, to achieve high-performance and low-power DRAM cells. However, this decrease in device size has led to increased geometric fluctuations during the semiconductor fabrication process, significantly impacting the retention time, which represents the performance of the DRAM device. To analyze the variations in retention time, Technology Computer-Aided Design (TCAD) simulation is commonly employed, taking into account these fluctuations. However, one major drawback is the lengthy simulation time. In this study, we propose an efficient approach to rapidly identify DRAM devices with short retention time among devices with geometric fluctuations. Specifically, we introduce three approaches: random sampling, machine learning (ML), and particle swarm optimization (PSO) algorithms, to enable efficient simulations. These approaches allow for effective research on leakage current and retention time distribution in DRAM devices with geometric fluctuations. Furthermore, we successfully identify and comprehensively analyze the worst-performing DRAM device.